Space2Ground is a multi-level, multi-sensor, multi-modal dataset, annotated with grassland/non-grassland labels for agriculture monitoring. We combine Sentinel-1 SAR data, Sentinel-2 multispectral data and street-level images for the purpose of grassland detection.
In particular, our dataset consists of:
- Space Component
- Sentinel-1 monthly mean time-series
- Sentinel-2 time-series
- Ground Component
- Street level image patches obtained from Mapillary after appropriate processing
- Labels/Annotations
- Parcel IDs and Grassland/Non-Grassland labels, according to farmers' declarations (parcel_annotations.csv file).
To download Space2Ground, you can use the following links, taken from the corresponding page of the dataset on Zenodo:
- Sentinel-1 and Sentinel-2 analysis ready data (download link)
- file size: ~120MB zipped -> ~300MB when unzipped
- Street-level image patches (download link)
- file size: ~190MB zipped -> ~210MB when unzipped
- (Optional) - Raw/Initial set of street-level images (download link)
- file size: ~4.6GB zipped -> ~4.7GB unzipped
- file size: ~4.6GB zipped -> ~4.7GB unzipped
An alternative way to download Space2Ground, is to use the download.sh script included in this repository. The instructions for the script usage are as follows:
- download Sentinel data to
data/(~ 115MB)
bash download.sh sentinel- download Street level patches to
data/streetLevel_patches(~ 187MB)
bash download.sh imagepatches- (Optional) download Street level raw images to
data/Mapillary(~ 4.6GB)
bash download.sh mapillaryor download everything with bash download.sh all
By downloading and unzipping the components, the file structure should look as follows:
├── Mapillary (optional)
│ └── mapillary
│ ├── 1000364310368104.jpg
│ ├── 1000636817135431.jpg
| ... (~ 38000 more images)
├── s1
│ ├── test_dataset.csv
│ └── train_dataset.csv
├── s2
│ ├── test_dataset.csv
│ └── train_dataset.csv
└── streetLevel_patches
├── test
│ ├── Grassland
│ │ ├── 1021749698360014_right.jpg
│ │ ├── 1076131292797553_left.jpg
│ │ ... (~ 1900 more images)
│ └── Non_Grassland
│ │ ├── 1003028817110323_right.jpg
│ │ ├── 1074590653065936_left.jpg
│ │ ... (~ 320 more images)
└── train
├── Grassland
│ ├── 1003976380345602_right.jpg
│ ├── 1005946389941606_left.jpg
│ ... (~ 6950 more images)
└── Non_Grassland
├── 1005486846651292_left.jpg
├── 1019023345502754_right.jpg
... (~ 900 more images)
The following table is a 2-row sample of the Sentinel-1 csv data columns and how they look like
| label | id | 2017-04_VH_SAR | 2017-04_VV_SAR | 2017-05_VH_SAR | 2017-05_VV_SAR | ... |
|---|---|---|---|---|---|---|
| 1 | 38267 | 3.880000039935112000e-01 | 3.217499963939189911e-01 | 3.055000031987826214e-01 | 2.696666667858759747e-01 | ... |
| 0 | 12914 | 1.291400000000000000e+04 | 3.467499986290931702e-01 | 2.702500000596046448e-01 | 2.591666653752326965e-01 | ... |
| ... |
The full list of column names of the Sentinel 1 data is:
s1_colnames = ['label', 'parcel_id', '2017-04_VH_SAR', '2017-04_VV_SAR', '2017-05_VH_SAR',
'2017-05_VV_SAR', '2017-06_VH_SAR', '2017-06_VV_SAR', '2017-07_VH_SAR',
'2017-07_VV_SAR', '2017-08_VH_SAR', '2017-08_VV_SAR', '2017-09_VH_SAR',
'2017-09_VV_SAR', '2017-10_VH_SAR', '2017-10_VV_SAR', '2017-04_VH_COH',
'2017-04_VV_COH', '2017-05_VH_COH', '2017-05_VV_COH', '2017-06_VH_COH',
'2017-06_VV_COH', '2017-07_VH_COH', '2017-07_VV_COH', '2017-08_VH_COH',
'2017-08_VV_COH', '2017-09_VH_COH', '2017-09_VV_COH', '2017-10_VH_COH',
'2017-10_VV_COH']The following table is a 2-row sample of the Sentinel-2 csv data columns and how they look like
| label | id | B02_2017-03-24 | B02_2017-03-27 | ... | B8A_2017-03-24 | ... |
|---|---|---|---|---|---|---|
| 1 | 12914 | 1.877940319825072701e-01 | 2.718666706611517214e-01 | ... | 3.002316555825695898e-01 | ... |
| 0 | 41660 | 1.720458268082157016e-01 | 2.467546732731773496e-01 | ... | 1.456669014803818607e-01 | ... |
| ... |
The full list of column names of the Sentinel 2 data is:
s2_colnames = ['label', 'id', 'B02_2017-03-24', 'B02_2017-03-27', 'B02_2017-04-26', 'B02_2017-05-06', 'B02_2017-05-26',
'B02_2017-06-02', 'B02_2017-06-05', 'B02_2017-07-07', 'B02_2017-07-17', 'B02_2017-07-20', 'B02_2017-07-22',
'B02_2017-08-04', 'B02_2017-08-06', 'B02_2017-08-09', 'B02_2017-08-14', 'B02_2017-08-19', 'B02_2017-08-24',
'B02_2017-08-26', 'B02_2017-08-29', 'B02_2017-09-03', 'B02_2017-09-10', 'B02_2017-09-15', 'B02_2017-09-20',
'B02_2017-09-23', 'B02_2017-09-28', 'B02_2017-10-03', 'B02_2017-10-05', 'B02_2017-10-15', 'B02_2017-10-18',
'B03_2017-03-24', 'B03_2017-03-27', 'B03_2017-04-26', 'B03_2017-05-06', 'B03_2017-05-26', 'B03_2017-06-02',
'B03_2017-06-05', 'B03_2017-07-07', 'B03_2017-07-17', 'B03_2017-07-20', 'B03_2017-07-22', 'B03_2017-08-04',
'B03_2017-08-06', 'B03_2017-08-09', 'B03_2017-08-14', 'B03_2017-08-19', 'B03_2017-08-24', 'B03_2017-08-26',
'B03_2017-08-29', 'B03_2017-09-03', 'B03_2017-09-10', 'B03_2017-09-15', 'B03_2017-09-20', 'B03_2017-09-23',
'B03_2017-09-28', 'B03_2017-10-03', 'B03_2017-10-05', 'B03_2017-10-15', 'B03_2017-10-18', 'B04_2017-03-24',
'B04_2017-03-27', 'B04_2017-04-26', 'B04_2017-05-06', 'B04_2017-05-26', 'B04_2017-06-02', 'B04_2017-06-05',
'B04_2017-07-07', 'B04_2017-07-17', 'B04_2017-07-20', 'B04_2017-07-22', 'B04_2017-08-04', 'B04_2017-08-06',
'B04_2017-08-09', 'B04_2017-08-14', 'B04_2017-08-19', 'B04_2017-08-24', 'B04_2017-08-26', 'B04_2017-08-29',
'B04_2017-09-03', 'B04_2017-09-10', 'B04_2017-09-15', 'B04_2017-09-20', 'B04_2017-09-23', 'B04_2017-09-28',
'B04_2017-10-03', 'B04_2017-10-05', 'B04_2017-10-15', 'B04_2017-10-18', 'B05_2017-03-24', 'B05_2017-03-27',
'B05_2017-04-26', 'B05_2017-05-06', 'B05_2017-05-26', 'B05_2017-06-02', 'B05_2017-06-05', 'B05_2017-07-07',
'B05_2017-07-17', 'B05_2017-07-20', 'B05_2017-07-22', 'B05_2017-08-04', 'B05_2017-08-06', 'B05_2017-08-09',
'B05_2017-08-14', 'B05_2017-08-19', 'B05_2017-08-24', 'B05_2017-08-26', 'B05_2017-08-29', 'B05_2017-09-03',
'B05_2017-09-10', 'B05_2017-09-15', 'B05_2017-09-20', 'B05_2017-09-23', 'B05_2017-09-28', 'B05_2017-10-03',
'B05_2017-10-05', 'B05_2017-10-15', 'B05_2017-10-18', 'B06_2017-03-24', 'B06_2017-03-27', 'B06_2017-04-26',
'B06_2017-05-06', 'B06_2017-05-26', 'B06_2017-06-02', 'B06_2017-06-05', 'B06_2017-07-07', 'B06_2017-07-17',
'B06_2017-07-20', 'B06_2017-07-22', 'B06_2017-08-04', 'B06_2017-08-06', 'B06_2017-08-09', 'B06_2017-08-14',
'B06_2017-08-19', 'B06_2017-08-24', 'B06_2017-08-26', 'B06_2017-08-29', 'B06_2017-09-03', 'B06_2017-09-10',
'B06_2017-09-15', 'B06_2017-09-20', 'B06_2017-09-23', 'B06_2017-09-28', 'B06_2017-10-03', 'B06_2017-10-05',
'B06_2017-10-15', 'B06_2017-10-18', 'B07_2017-03-24', 'B07_2017-03-27', 'B07_2017-04-26', 'B07_2017-05-06',
'B07_2017-05-26', 'B07_2017-06-02', 'B07_2017-06-05', 'B07_2017-07-07', 'B07_2017-07-17', 'B07_2017-07-20',
'B07_2017-07-22', 'B07_2017-08-04', 'B07_2017-08-06', 'B07_2017-08-09', 'B07_2017-08-14', 'B07_2017-08-19',
'B07_2017-08-24', 'B07_2017-08-26', 'B07_2017-08-29', 'B07_2017-09-03', 'B07_2017-09-10', 'B07_2017-09-15',
'B07_2017-09-20', 'B07_2017-09-23', 'B07_2017-09-28', 'B07_2017-10-03', 'B07_2017-10-05', 'B07_2017-10-15',
'B07_2017-10-18', 'B08_2017-03-24', 'B08_2017-03-27', 'B08_2017-04-26', 'B08_2017-05-06', 'B08_2017-05-26',
'B08_2017-06-02', 'B08_2017-06-05', 'B08_2017-07-07', 'B08_2017-07-17', 'B08_2017-07-20', 'B08_2017-07-22',
'B08_2017-08-04', 'B08_2017-08-06', 'B08_2017-08-09', 'B08_2017-08-14', 'B08_2017-08-19', 'B08_2017-08-24',
'B08_2017-08-26', 'B08_2017-08-29', 'B08_2017-09-03', 'B08_2017-09-10', 'B08_2017-09-15', 'B08_2017-09-20',
'B08_2017-09-23', 'B08_2017-09-28', 'B08_2017-10-03', 'B08_2017-10-05', 'B08_2017-10-15', 'B08_2017-10-18',
'B8A_2017-03-24', 'B8A_2017-03-27', 'B8A_2017-04-26', 'B8A_2017-05-06', 'B8A_2017-05-26', 'B8A_2017-06-02',
'B8A_2017-06-05', 'B8A_2017-07-07', 'B8A_2017-07-17', 'B8A_2017-07-20', 'B8A_2017-07-22', 'B8A_2017-08-04',
'B8A_2017-08-06', 'B8A_2017-08-09', 'B8A_2017-08-14', 'B8A_2017-08-19', 'B8A_2017-08-24', 'B8A_2017-08-26',
'B8A_2017-08-29', 'B8A_2017-09-03', 'B8A_2017-09-10', 'B8A_2017-09-15', 'B8A_2017-09-20', 'B8A_2017-09-23',
'B8A_2017-09-28', 'B8A_2017-10-03', 'B8A_2017-10-05', 'B8A_2017-10-15', 'B8A_2017-10-18', 'B11_2017-03-24',
'B11_2017-03-27', 'B11_2017-04-26', 'B11_2017-05-06', 'B11_2017-05-26', 'B11_2017-06-02', 'B11_2017-06-05',
'B11_2017-07-07', 'B11_2017-07-17', 'B11_2017-07-20', 'B11_2017-07-22', 'B11_2017-08-04', 'B11_2017-08-06',
'B11_2017-08-09', 'B11_2017-08-14', 'B11_2017-08-19', 'B11_2017-08-24', 'B11_2017-08-26', 'B11_2017-08-29',
'B11_2017-09-03', 'B11_2017-09-10', 'B11_2017-09-15', 'B11_2017-09-20', 'B11_2017-09-23', 'B11_2017-09-28',
'B11_2017-10-03', 'B11_2017-10-05', 'B11_2017-10-15', 'B11_2017-10-18', 'B12_2017-03-24', 'B12_2017-03-27',
'B12_2017-04-26', 'B12_2017-05-06', 'B12_2017-05-26', 'B12_2017-06-02', 'B12_2017-06-05', 'B12_2017-07-07',
'B12_2017-07-17', 'B12_2017-07-20', 'B12_2017-07-22', 'B12_2017-08-04', 'B12_2017-08-06', 'B12_2017-08-09',
'B12_2017-08-14', 'B12_2017-08-19', 'B12_2017-08-24', 'B12_2017-08-26', 'B12_2017-08-29', 'B12_2017-09-03',
'B12_2017-09-10', 'B12_2017-09-15', 'B12_2017-09-20', 'B12_2017-09-23', 'B12_2017-09-28', 'B12_2017-10-03',
'B12_2017-10-05', 'B12_2017-10-15', 'B12_2017-10-18']These are cropped vegetation parts of mapillary street-level images. The methodology of creating these images is described in our Space2Ground paper.
- The label of each image (Grassland/Non-Grassland) can be derived either
- by the folder in which it belongs to (see Dataset Structure section)
- by the mappings provided in the parcel_annotations.csv file
- The
leftandrightparts of each filename correspond to the information related to the side of the image that the parcel is located. This information is useful for the front-facing acquisitions, where the road ahead separates the parcels on the left and right.
Samples of Grassland patches:
Samples of Non-Grassland patches:
This is a csv file that contains the annotations that connect all the Space2Ground dataset components, as it holds information on the parcel id and the street level image id. Its structure is as follows:
| id | label | image_date | image_id | direction |
|---|---|---|---|---|
| 1890 | Grassland | 2017-08-02 | 527903208225456 | right |
| 28235 | Grassland | 2017-08-05 | 485698512854670 | left |
| 34666 | Maize | 2017-08-13 | 306183544216406 | left |
| ... |
Note that the labels include crop types, like Grassland, Maize, Onions, etc. The Grassland/Non-Grassland differentiation has been generated by processing the labels accordingly.
A jupyter notebook with the implementation using Space2Ground dataset described in our publication can be also found as part of this repo (file Space2Ground_use_case.ipynb)
The process we followed to extract the vegetation patches from the "raw" Mapillary dataset can be found in this jupyter notebook. A new, detailed version with further instructions and comment will be uploaded shortly.
-
Download street level images from Mapillary and annotate them with labels from openly accessible Land Parcel Identification Systems (LPIS) data https://github.com/Agri-Hub/Callisto/tree/main/Mapillary
-
Download and pre-process Sentinel-1 and Sentinel-2 data: https://github.com/Agri-Hub/ADC
[1]
George Choumos, Alkiviadis Koukos, Vasileios Sitokonstantinou and Charalampos Kontoes (2022)
Towards Space-to-Ground Data Availability for Agriculture Monitoring
2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP 2022), 26-29 June 2022, Nafplio, Greece
https://arxiv.org/abs/2205.07721
If you used Space2Ground dataset, please cite the following paper.
@INPROCEEDINGS{9816335,
author={Choumos, George and Koukos, Alkiviadis and Sitokonstantinou, Vasileios and Kontoes, Charalampos},
booktitle={2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP)},
title={Towards Space-to-Ground Data Availability for Agriculture Monitoring},
year={2022},
volume={},
number={},
pages={1-5},
doi={10.1109/IVMSP54334.2022.9816335}
}